A Dataset of EEG Signals Reflecting Learners’ Interest States

Learning interest is widely regarded as a key factor influencing learning outcomes; however, its underlying neural mechanisms remain insufficiently understood. In this study, we constructed an EEG dataset collected in realistic learning scenarios to investigate the neural representations of learning interest. To capture interest states under two contrasting conditions—autonomy support and autonomy deprivation—we designed two EEG acquisition paradigms: an intrinsic-interest experiment and an extrinsically assigned-interest experiment. A total of 26 participants were recruited and divided into two groups. Each group completed the learning tasks according to one of the paradigms, while 16-channel EEG signals were synchronously recorded throughout the learning process. The dataset includes power features and peak-frequency indices across five canonical frequency bands. To validate the usability of the dataset, we applied five classical machine-learning models—logistic regression, support vector machine (SVM), decision tree, k-nearest neighbors (k-NN), and naive Bayes—to classify interest states, achieving mean accuracies of 98.32% and 96.42% for the intrinsic-interest and extrinsically assigned-interest datasets, respectively. This dataset provides a valuable resource for research in educational neuroscience, learning and cognitive mechanisms, and brain–computer interface applications.